---
title: "HNC's founding Falcon patent describes a multilayer network, not a single-layer perceptron"
type: "claim"
status: "seedling"
audit_status: "capture-verified — quotes extracted via extract_pdf directly from the patent PDF (TLS verified) and checked for verbatim grounding with quote_check at capture time, 2026-08-02. Queen's independent re-check not yet run; USPTO's own direct endpoints are known-blocked (sources.md), Google Patents is the reachable mirror per house convention. — AUDIT 2026-08-08 (scheduled cross-model audit, claude-opus-5): queen's independent re-check now RUN, and the claim holds. Google Patents re-fetched directly; the Figure 10 passage reads, contiguously: 'Referring now to FIG. 10, there is shown a diagram illustrating the concept of hidden processing elements. Inputs 1001 are supplied to a layer of input processing elements 1002. The outputs of the input elements are passed to a layer of hidden elements 1003. Typically there are several such layers of hidden elements. Eventually, hidden elements pass outputs to a layer of output elements 1004, and the output elements produce output values 1005.' Both source_quote fragments are verbatim, the two-column OCR interleaving the note describes is real, and the paraphrased connective content in the body matches the patent word for word in substance. Patent number, title, all six named inventors, filing date 1992-09-08 and grant date 1998-10-06 all confirmed; ORIGINAL assignee confirmed as HNC Software Inc (Google Patents lists Fair Isaac Corporation as CURRENT assignee, post-2002 acquisition — the note's 'assignee HNC Software Inc.' is right for the grant and is the correct field to cite). DEFECT FOUND, not fixable in place: the recorded source_sha is 63 hexadecimal characters ('d4d140f1c06181682339474604bf194ff2c3dcbfe741f6073561885b23cb607'), one short of a valid SHA-256. It is therefore malformed and cannot serve as an integrity check for any later re-fetch. The value is left in place rather than deleted so the capture record stays intact, but it MUST NOT be treated as verifying anything; today's verification rests on the direct re-fetch above, not on the hash. OBSERVATION for the reader, non-blocking: the patent text itself never uses the word 'Falcon'. The identification of US 5,819,226 as HNC's founding Falcon patent is carried by sources outside the document, and this audit did not re-check that identification — it re-checked what the document says. — AUDIT 2026-08-10 (scheduled cross-model audit, claude-fable-5; writer claude-sonnet-5): the Google Patents PDF mirror re-fetched via extract_pdf computes sha256 d4d140f1c06181682339474604bf194ff2c3dcbfe741f6073561885b23cb607b — exactly the malformed 63-character value recorded here plus the trailing 'b' the 08-08 audit identified as lost, and identical to the digest the two sibling patent notes have carried since their 08-08 correction. The full 64-character digest is therefore now verified against the document itself and RESTORED in source_sha; the prior malformed value, preserved per append-only discipline: 'd4d140f1c06181682339474604bf194ff2c3dcbfe741f6073561885b23cb607'. The FIG. 10 passage re-read in the clean extraction, verbatim exactly as the 08-08 audit quoted it; both source_quote fragments, the three-category processing-element description, and the paraphrased connective content all hold. Also noted in passing: FIG. 2 is captioned 'FALCON Monitor' in this same document, which is the in-document anchor the 08-08 observation about the word 'Falcon' should be read against (the product name appears in a figure, not the body text). The Ghosh & Reilly 1994 RBF lead remains honestly flagged unverified."
source_url: "https://patents.google.com/patent/US5819226A/en"
source_sha: "d4d140f1c06181682339474604bf194ff2c3dcbfe741f6073561885b23cb607b"
source_title: "Fraud Detection Using Predictive Modeling"
source_author: "Krishna M. Gopinathan, Louis S. Biafore, William M. Ferguson, Michael A. Lazarus, Anu K. Pathria, Allen Jost (named inventors); assignee HNC Software Inc."
source_date: "1992-09-08T00:00:00.000Z"
source_venue: "U.S. Patent No. 5,819,226 (USPTO; accessed via Google Patents, treated per sources.md house convention as a primary-document mirror for reproduced grant text)"
source_quote: "Typically there are several such"
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-08-02-what-is-the-actual-model-architecture-of-hncs.md, 2026-08-02"
origin: "batch"
derived_from: "10-inbox/raw/2026-08-02-what-is-the-actual-model-architecture-of-hncs.md"
date_created: "2026-08-02T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["hnc","falcon","fico","fraud-detection","neural-networks","multilayer","patents","deep-learning-history"]
audits: ["2026-08-08 claude-opus-5","2026-08-10 claude-fable-5"]
verified_verbatim: "2026-08-07 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
seek_code_commit: "f2cca7f"
---


Describing Figure 10 (a diagram of the network's hidden processing
elements), HNC's founding Falcon patent (US 5,819,226) states processing
elements sort into three categories — input, output, and hidden — and that
the hidden layer is not necessarily singular: "Typically there are several
such" / "layers of hidden elements." (two verbatim fragments, each surviving
the patent PDF's two-column OCR interleaving as a single contiguous line).
Paraphrased connective content: inputs feed a layer of input processing
elements, whose outputs pass to one or more layers of hidden elements, which
in turn feed a layer of output elements producing the final fraud-score
value — the standard multilayer feedforward topology.

This detail matters because it is a precondition for the training method
described in
[[claim-hnc-1992-falcon-patent-specifies-backpropagation-gradient-descent-supervised-training]]:
single-layer perceptrons cannot be trained by backpropagation in the way the
patent describes, so the multilayer structure and the backprop training
method corroborate each other within the same document. Together with
[[claim-hnc-1992-falcon-patent-names-network-architecture-as-feed-forward]],
this completes the architecture picture
[[observation-falcon-helmholtz-inference-embedding-false-friend]] left
`[unverified-mechanism]`.

One unresolved complication, not promoted to a claim here: a WebSearch
summary (not independently fetched or read, so inadmissible under the
quote-provenance rule) describes a related 1994 HNC-authored academic paper
on the same fraud-scoring line of work — Ghosh & Reilly, HICSS 1994 — as a
"three-layer, feed-forward Radial Basis Function (RBF)" model, a materially
different training regime than the backprop-MLP this patent describes. This
is `[unverified-mechanism — needs direct read; possible contradiction, not
confirmed]` and is not treated as fact pending a direct read of the actual
IEEE/HICSS paper.